If they actually did this correctly, it would be great. Whether or not it's possible, or even desirable to eliminate all hate speech, it should be possible to minimize the harms.
When somebody mutters some hateful comment to themselves, do we care? Not really. We care that the hateful comment gets repeated and amplified. We care that someone might take that hateful comment. We care that someone might take harmful actions based on the comment.
If those algorithms successfully let these comments die in ignominy they've done their job. My fear is that they won't really do this though. Instead they'll mostly hide these comments and "accidentally" let them slip out when the company thinks they need an "engagement" boost.